Software & Systems

Articles of journal № 3 at 2012 year.

Order result by: Public date | Title | Authors

11. Data integration and query language in large information infrastructures [№3 за 2012 год]
Authors: Коваленко В.Н., Куликов А.Ю.
Visitors: 6745
Automation of various forms of professional activity by means of computer technologies generates arrays of information stored in databases. This information is used, first, inside organizations, but can be helpful for solving important tasks outside their borders. Development of the corresponding applications is considerably complicated in the absence of the specialized system tools supporting access to data from multiple databases. In this direction, named data integration, the general, application-independent methods, allowing consolidation of heterogeneous databases are developed. The tools created on such basis are used in practice, however the problem of their scalability on the number of integrated databases remains open. In the paper, an approach to the problem of mass integration (tens and hundred databases) is described. The two questions are considered that seems to be the most essential under these conditions: the method of data integration and the type of informational queries. The integration method allows defining a representation (the global scheme) in which the data of the integrated databases form a common unified space. The method is aimed at the creation of information infrastructures with dynamically changing database set: change of a set does not require modification of the global scheme or existing applications. The language of search queries is an SQL-92 extension, with the difference that the operations are executed on subsets of databases. In addition, databases are not addressed explicitly: descriptive information – meta-attributes – is used for their se-lection. Such type of queries allows creating applications capable of processing data from varied sets of sources.

12. Study of approximative possibilities of radial-basic network with orthogonal polynoms [№3 за 2012 год]
Author: I.V. Lyozina
Visitors: 7899
The article describes set up of the problem of approximation, provides rationalization for possibility to use the radial-basic neural network as an approximator of probability density function, gives the approximating statement for this network and statement for objective function, by means of which selection of basic function parameters and weight values is done. Besides, this article considers the possibility to use not only canonical Gaussian functions in the course of approximation of probability density function with a radial-basic network, but also sigmoidal functions, power functions and orthogonal polynoms of Legendre, Chebyshev of I and II kind, Laguerre and Hermite, and gives the relevant formulae. It compares the accuracy of approximation by means of computation of mean-square deviation. Approximation of probability density of Simpson and Rayleigh with a radial-basic network with sigmoidal, power functions as well as polynoms of Legendre, Chebyshev of I and II kind, Laguerre and Hermite are given as examples. At the end there is a recommendation to use the radial-basic network with polynoms of Legendre, Chebyshev of I and II kind as basic functions in case of increase of neurons number in the buried layer, because this network allows lower values of mean-square deviation than the one with canonical Gaussian functions to be achieved.

13. Combined method of text tonality automatic identifying [№3 за 2012 год]
Author: Котельников Е.В.
Visitors: 12321
The article is devoted to the problem of automatic text sentiment analysis. Sentiment analysis is used for many important applications, for example, the research for a commercial organization of the relations of customers to its production, or the development of a recommendatory system for the customers of specified groups of goods or services. The new method of text sentiment analysis which combines the results of work of Support Vector Machines (SVM) and Keywords method is suggested in the article. The SVM is a well-known classification method which shows good results on text analysis. The Keywords method is based on the counting of weights of features which the text contains. Words and word combinations, the importance of which is evaluated on the base of Relevance Frequency (RF), are used as features. The combined method contains two phases – training and recognition. On the first phase SVM-classifier and Keywords-classifier are trained on the labeled collection of texts. On the second phase the results of recognition by both classifiers for a new text are combined and the final decision is formed; confidence levels of both classifiers are taken into account. Each of the used methods (SVM, Keyword, combined methods) has its set of parameters. The selection of optimal values of parameters is accomplished with the help of cross-validation. The effectiveness of the combined method is proved experimentally on the text collections of seminar ROMIP-2011. The metrics of precision, recall, accuracy and F1 measure are used for the assessment. F1 measure of the suggested method is equal to 79 %, what is two per cent more than the best result achieved on the seminar ROMIP-2011.

14. Computing and modeling of fuzzy Cauchy problem using virtual perspective method [№3 за 2012 год]
Author: Myshev A.V.
Visitors: 8061
The work describes computing and modeling of fuzzy Cauchy problem using virtual perspective method with limitations, closedness and indeterminancy. It was investigated dynamic evolution of space objects within predicative model of bounded problem N – mass body using methods of standard computational mathematics and virtual perspective method. Obtain results showed that virtual perspective method was more effective and advanced in comparison with traditional simulation models. Computational modeling of this problem using virtual perspective method it is described on information modeling language. In this case the problem is defined and set as a discrete dynamic information system, specified at the set of nodes of coordinate and perspective arrays. Copmputing systems present character strings and its compositions as a information prototype of computing environment and its elements. Modeling in the virtual environment is characterized by following specifics. First, computation process creates dynamics of related image links (RIL) in modeling technologies. Second, information interaction mode and relations of the virtual objects in the model environment in the form of RIL conditions and real conditions of simulated problem, is reflected in coordinate nodes. This array is a geometry pattern of active memory of the computation environment. It was used two types of discrete sampling of the formal evolution operators for the modeled problem; they were used for RIL construction that corresponds to dynamics equation of the Cauchy problem modeling. Two patterns for RIL construction were shown in computing modeling of the fuzzy Cauchy problem. First model was constructed using Runge-Kutta techniques, and the second one was constructed using Adams techniques. Currently, in computing modeling technologies these methods were studied and examined best of all. For the fuzzy problems, related image links according to any variable of the evolution operator were considered as set of projections of the local information dynamic system in the basic array nodes.

15. Mathematic model of the user Network-on-Chip [№3 за 2012 год]
Authors: Mosin S.G., Хассан Мд Муид, Тухтамирзаев А.Ю.
Visitors: 7976
Microelectronics development provided implementation of complex electronic systems in integrated-circuit form. CAD systems, integral technologies and design cycle during design of electronic system are selected to reduce time of their creation, increase reliability and quality of the final product. The effectiveness of the design process can be improved with different approaches, e.g. SoC – system on a chip, SiP – system in a package, MCM – multi chip module etc. For design of complex MPSoC – Multiprocessors System on a Chip it was provided NoC technology – network on a chip. Architectures of specialized MPSoC applications include numerous heterogeneous computing cores and memory modules. Each core provides a limited set of applicable functions. Such projects can be provided with simple inter-core communication. NoC technology is used for design of communication environment that provides communication between different system modules. NoC consists of routers that are physically connected with each other. Each computation core and memory blocks are connected to NoC by means of interface to the network (RNI-interface). In general case, NoC technology suggests the use of homogeneous topology – the grid, which provides connection of the same number of cores to each switch that form the domain. Domains in the system interact and they represent a regular structure. Alternative solution presents usage of heterogeneous topology; this fact assumes consideration of specific core features during design stage of the designed system. Number of routers and switching method are selected with the purpose of minimization of signal latency, chip size and power consumption. Heterogeneous topology helps to design specialized applications with minimal versatile properties. There is provided mathematical model of the user network-on-chip (NoC). There is provided search shortest path in the flow chart. Obtained outputs of algorithm operation for NoC topology optimization are provided.

16. Metadescriptions and cataloguing of scientific information resources of the RAS [№3 за 2012 год]
Authors: Еркимбаев А.О., A.B. Zhizhchenko, Зицерман В.Ю., Кобзев Г.А., Serebryakov V.A., A.N. Sotnikov, Шиолашвили Л.Н.
Visitors: 10443
A large part of scientific knowledge is formalized in the form of electronic resources – data and knowledge bases, electronic reference books, etc. Work with electronic resources, including their adaptation to the subject area, systematization and accumulation of data, achieved an equal status with theory and experiment. There appeared such subjects as bio- and geoinformatics, which subject of study is submission of complex data. However, with the spreading of databases and similar means deep problems arose caused by lack of interoperability. Autonomy of resources functioning, diversity of data formats and structures, lack of data presentation standards – not all the reasons complicating the data exchange. In the global and domestic practice in recent years there have been selected approaches to possible resolving problems using versions of XML language, for standardizing of metadata system and terms dictionaries within a certain area of expertise, such as CML versions for submission of chemical data, MatML – for material science, ThermoML – for thermodynamics. An insistent in elaborating principles and technologies for integration of many RAS resources led to the formation of an extensive program on creation of so-called Data Centre. It is expected that this project will help to overcome the fragmentation and limited availability of digital resources in the form of databases, electronic publications, data-processing tools, supported by various institutes of the Russian Academy of Sciences. In this work as the first phase of the integration is offered the system of resources certification, adequately reflecting the subject area, resource types, access conditions, etc. A portal is developed on which there is an extensive set of metadata for each registered resource.

17. Method of decision-making support on management of temporary aspects of projects at the industrial enterprises [№3 за 2012 год]
Authors: Stoyanova O.V., Васицына А.И.
Visitors: 11908
This article presents a method of decision-making support for managing time characteristics of projects implemented in the industry. This method is based on the construction and use of temporal-logic model of the project, which allows a formal description of qualitative temporal relations between the different phases of the project. We describe the algorithms and tools used in the process of constructing this model, the algorithms of calculation time characteristics of projects and the possibility of using the results of simulation for decision support. It is given a matrix selection of the types of temporary structures and methods of its description depending on the characteristics of the project, the need for modeling instant events, types of relations between the stages and other factors. It is shown that for complex projects is the preferred choice of matter-temporal logic for interval or dot-interval-time structures. Features of each of these temporary structures taken into account in the algorithms for constructing models and formulas of the calculation of its parameters. Presents the results of simulation of computational experiments aimed at the investigation of the effectiveness of the work of the method for solving various tasks of decision-making support in the areas of: planning the project's structure, optimization of its duration, and manageability. It is shown that to obtain satisfactory results it is necessary to pay special attention to the preparation of describing the relationships between the stages of the project in terms of the use of temporal logic, i.e. the work in the initial stages of the proposed method. This can be used by well-known instruments, with a view of the marked in the work of the peculiarities of their application to the task. In general, the method can be recommended as a tool to support decision-making for management of projects with complex stationary structure. For projects with non-stationary structure of the method can be used for the construction and analysis of models of the project at selected intervals, which assumes the truth of the conditions of invariance of its structure.

18. Techniques provided fof valnerabilities reduction in special real time software [№3 за 2012 год]
Author: Нархов К.Г.
Visitors: 12935
This article is dedicated to automation of a programmer’s work, particularly in the area of vulnerabilities reduction and bugs in the program code. The article reviews details of the program design in technical facilities of automated code generation for special software (TFACG SS) and use of library contained in TFACG SS that reduces potential vulnerabilities appearing in new programs. The author provides taxonomy of typical vulnerabilities in real time programs, he makes review of each class of vulnerability, how often and why does the vulnerability occur, and its prevention using TFACG SS facilities. Some potential vulnerabilities take in account configuration of real time operating system. Taxonomy of vulnerabilities was made with a static estimating device and set of real time source code designed by Institute for Scientific Research of Russian Academy of Sciences. Set of the source codes includes 204 of the program modules (more than 111700 lines). The article finishes with example of reduction of potential vulnerabilities in real time source code generation program (PVSC RT), which is a part of TFACG SS. The article shows method of reduction of vulnerabilities using standard program patterns provided by TFACG SS, this method repaired all vulnerabilities that were found by static estimating device in PVSC RT. The library expansion for standard TFACG SS patterns and the supplementation of the static analyzer rules, which will include tests and conditions specific for real time operating system, are the main prospects of solutions presented in the article.

19. Microbenchmarks for microprocessor RTL-models performance assessment [№3 за 2012 год]
Author: Николина Н.В.
Visitors: 6110
An approach for evaluating and monitoring performance of the microprocessor on design stage is considered. The technique to estimate performance of separate blocks is offered, any potential influence of others is thus ignored. We propose a test suite for evaluating performance of microprocessor RTL-models (Register transfer level). The test set consists of the short programs (microtestbenches) directed on performance evaluating of separate blocks. A choice of test set for different modules is realized on taking into account features of its work. An article presents a number of test situations for evaluating such modules as instruction fetch and dispatch buffer (IFDB), floating point unit (FPU) and memory management unit (MMU) for MIPS-like architecture. For the analysis of run time performance counters are used, which are the parts of control coprocessor registers of microprocessor. Automation for creating test cases, regression performance measurement and visualization of performance evaluation results are proposed. During reasonable time the test system allows to receive results of a performance evaluation and to compare it with results of the previous versions of RTL model, or with reference values. Also the impact of performance measurements on architecture of future chip is considered. The possibility to investigate the influence on microprocessor performance of such factors as changing memory frequency is shown. The results of measurements are shown on example of performance evaluation of superscalar microprocessor which is developed in SRISA RAS. The results were confirmed on final crystal.

20. Multifunctional simulator of neural networks [№3 за 2012 год]
Author: Солдатова О.П.
Visitors: 13905
The article describes the neurosimulator implementing models of multilayered perceptron, radial-basic networks and fuzzy neural networks. It studies the efficiency of usage of neurosimulator to solve classification and forecasting problems. The article pays attention to the analysis of hybrid models of neural networks and fuzzy inference systems based on condition-action rules «if-then». It reviews the knowledge bases in fuzzy inference system of Mamdani-Zade, models of fuzzy inference TKS, Tsukamoto model, Wang-Mendel model of fuzzy production network. Based on the specified models the software complex is implemented, which makes it possible to check the accuracy of computations in the course of solution of classification and forecasting problems. Comparison of the results was carried out using the same data, in several series with different values of neural network parameters in order to identify the average error value at each type of problems. Comparison of accuracy of the obtained result was carried out as per the error function value, mean-square error and percentage error. In the course of solution of classification problems the best results were obtained in case of Wang-Mendel network and TKS. In the course of study of the efficiency of forecasting problem solutions the best results were shown by the networks implementing Mamdani-Zade and Tsukamoto models. This example clearly demonstrates impossibility to choose a single optimal model and for the problems of different classes one should use different models, which are to the best advantage applicable under the particular problem conditions.

◄ ← Preview | 1 | 2 | 3 | 4 | 5 | 6 | Next → ►